Population aging and housing prices: who are we calling old? · of potential economic effect of population aging, given ever-increasing life expectancy. This paper introduces alternatively

NBP Working Paper No. 288

Population aging and housing prices: who are we calling old?

Ye Jin Heo

Economic Research DepartmentWarsaw 2018

NBP Working Paper No. 288

Population aging and housing prices: who are we calling old?

Ye Jin Heo

Published by: Narodowy Bank Polski Education & Publishing Department ul. Świętokrzyska 11/21 00-919 Warszawa, Poland www.nbp.pl

ISSN 2084-624X

© Copyright Narodowy Bank Polski 2018

Ye Jin Heo – Graduate Institute of International and Development Studies; [email protected]

The views expressed herein are those of the author and not necessarily those of Narodowy Bank Polski.

3NBP Working Paper No. 288

ContentsAbstract 41 Introduction 52 Literature Review 83 Data and Measurement 11

3.1 Data Sources 113.2 Measures of Population Aging 13

4 Empirical Analysis 194.1 Empirical Strategy 194.2 Results 214.3 Robustness Checks 25

4.3.1 Robustness to alternative population measures 254.3.2 Robustness to old-age dependency ratio 274.3.3 Robustness to first-difference model 28

5 Projection under Forward-Looking Scenario 305.1 Population Projection 305.2 Real House Price Prediction 32

6 Conclusion 37References 38Appendix A 41Appendix B 42Appendix C 43Appendix D 45

Narodowy Bank Polski4

Abstract

Abstract

This paper empirically studies through which channel - between short expected

remaining life and withdrawal from the labor market - population aging affects real

house price more and how the effect can vary if old-age population is defined alter-

natively in a way to reflect different aspects of aging, using a panel data of OECD

countries. It finds that the main driver of a negative relationship between aging and

real house price comes from the later stage of life and not immediately after the age

of 65 or retirement. It also shows that the effective retirement age matters more in

explaining the relationship between aging and real house price than the age 65, since

the share of retired population has a nonlinear effect on real house price, while the

standard old-age population aged over 65 does not. When I project future real house

price, the standard old-age population only predicts a further decrease in real house

price as aging continues, whereas the retired population captures a positive marginal

effect and leaves room for policy intervention.

Keywords: Aging, house prices, demographics

JEL Classification Codes: J11, G12, R21

2


Chapter 1

1 Introduction

This paper empirically studies the relationship between real house price and populationaging in 22 OECD countries. OECD countries are aging fast, and its potential negativeeconomic effect has been much discussed in both academic and policy forums, such as inGreenspan (2003), Bloom et al. (2011), Liu and Spiegel (2011) and Yoon et al. (2014).However, there has been little discussion about whether people aged 65 or above, the mostcommonly used definition of old-age population, is the most suitable one for the analysisof potential economic effect of population aging, given ever-increasing life expectancy.This paper introduces alternatively defined old-age population based on different aspectsof population aging and compares how their effect on real house price differs from eachother.

5

15

25

% o

f popula

tion

65

75

85

Age

1970 1980 1990 2000 2010Year

Life expectancy Share of old aged 65+

Figure 1: Median life expectancy and the share of old

Figure 1 shows how life expectancy (unit on the left axis) and the share of people aged65 or above (unit on the right axis) had evolved between 1970 and 2014 in 22 OECDcountries. Life expectancy, and therefore, the share of people aged 65 or above has beenconsistently increasing over time; in the beginning of the sample period, the median lifeexpectancy is 71.4 years and the median share of standard old-age people is 10.6%, but

3


it increased to 81.3 years and 18.2%, respectively, in 2014, which confirms that popula-tion is aging in OECD countries. However, this definition of old gives little informationon how age-specific behavior within this group may have changed as life expectancy in-creases.

I investigate how real house price is associated with population aging by focusing onthe hypothesis that two different aspects of aging - a distance to life expectancy and laborforce participation - may affect housing demand of older population, which consequentlyalso may affect real house price. Thus, this paper introduces two alternatively definedold-age population based on a distance to life expectancy and the effective retirementage.1 Since life expectancy and the effective retirement age change over time and acrosscountries, unlike the standard fixed age 65, the alternatively defined old-age populationonly includes either people whose expected remaining life is short or people who areeffectively withdrawn from labor force in each year. It can be useful to define old in acountry- and time-specific way since today’s old people are likely to behave differentlyfrom old who lived 40 years ago, even if they are the same age, as their expected lifehorizon is much longer now than their counterparts.

To be specific, I first regress real house price on the share of standard old-age popu-lation as a benchmark, and then I regress real house price on the share of old defined bya distance to life expectancy and the effective retirement age, respectively. By doing so,this paper answers the following two questions: First, how does increase in the share ofstandard old-age population affect real house price in a panel setting? Second, how doesthe effect of old-age population change if old is redefined in a way to reflect differentaspects of aging, namely expected remaining life and retirement?

Old-age population can negatively affect house prices through the lack of housing de-mand. It is likely that older people already own their home, therefore demand is weakerfor them to buy a new house compared to the working-age population. Also, they maysell off a house and move to a smaller one or change to a rented house after retirement tofinance consumption for the rest of their lives, if they bought a house as investment. Thisnegative relationship between population aging and house price has been much discussedin previous studies. However, if life expectancy increases fast enough so that their ex-pected remaining life is still relatively long even if they are retired, then older people may

1The effective retirement age is the average age of all persons withdrawing from the labor force in agiven period, calculated by the OECD. It is explained more in detail in Section 2, where the alternativedefinitions of old-age population is discussed.

4


Introduction

postpone selling their houses or even buy a house. This thought experiment suggests thatthe effect of old-age population on real house price may differ, depending on whether oldis defined by expected remaining life or by labor force participation.

This study has three main findings. First, it provides empirical evidence that increasein the share of elderly population is negatively related to real house price: 1 percentpoint increase in the share of elderly population is associated with between 4% to 6%decrease in real house price, subject to the definition of old-age population. It shows thatthe magnitude of negative effect is largest when old is defined as people whose expectedremaining life is ten years or below, which seems to suggest that the negative effect ofaging on real house price is driven by the expected remaining years of life rather thanretirement. Finally, it finds that explaining the relationship between aging and real houseprice by distinguishing adult population between “effectively” retired and those not makesmore economic sense, than using the arbitrary age 65 to define old-age population: anonlinear effect of aging on real house price is detected only when old is defined aseffectively retired population.

The rest of the paper is organized as follows. Section 2 reviews the relevant literature,and Section 3 describes data and alternative definitions of old-age population. Section 4presents the empirical strategy, results and interpretations. Section 5 estimates future realhouse prices in selected countries and discusses the result, and Section 6 concludes thepaper.

5


Chapter 2

2 Literature Review

Since the relationship between aging and the savings behavior was pioneered by Modiglianiand Brumberg (1954), studies on the effect of demographic changes have been based onthe life-cycle theory of savings. Although the life-cycle hypothesis predicts that agentsdissave when they are old, the hypothesis is not always strongly supported by empiricalevidence, especially when microeconomic data is employed (see Poterba (1994), Deatonand Paxson (1997), Hildebrand (2001), Battistin et al. (2009), Nardi et al. (2015)). Inthe literature of demographic effect on asset market, so-called “demographic doomsday”scenario, which predicts a large adverse effect of the aging population on the asset mar-ket, has been the subject of debates generating two opposite views in the literature: thosewho agree with the non-negligible negative demographic effect (for example, Mankiwand Weil (1989), Yoo (1994), Takats (2010), and Liu and Spiegel (2011)) and those whobelieve that demographic change is likely to bring only a limited effect, if any (for exam-ple, Poterba (2001), Abel (2003), and Davis and Li (2003)). Although the “asset-marketmeltdown” prediction by Mankiw and Weil (1989) in the 1990s is proved wrong, demo-graphic effect on asset market is still a topic that has mixed views in the literature.

This study contributes to several strands of the literature. The current literature onthe effect of population aging on house price has three limitations. First, commonlyused measures of aging in the literature are not sufficient to examine potentially changingimplications of population aging over time. By fixing the threshold age as 65, regardlessof the speed of aging in each country, it explores little about heterogeneity among elderlypeople who are aged more than 65. Second, given the amount of studies on determinantsof real house price, there is relatively less empirical work that exclusively studies the roleof population aging. Third, many empirical papers look at population aging in a single-country setting, which does not incorporate possible consequences of aging that mightalready be happening in other fast-aging countries. This paper attempts to address eachof these three points.

First of all, it introduces three different measures of old-age population. A numberof papers study the relationship between demographic changes and macroeconomic vari-ables, but most of them focus on certain “age” groups to capture population aging. Forinstance, Yoo (1994), Davis and Li (2003), Goyal (2004), Poterba (2004), and Brooks(2006) use the size of detailed adult age groups (i.e., population aged 25-34, 35-44, ...,65+) or 5-year age groups (i.e. population aged 0-4, 5-9, ..., 65+), while Nishimura and

6


Literature Review

Takats (2012) uses the size of working-age population (i.e. population aged 20-64) andBakshi and Chen (1994) uses the average age of total population. On the other hand,the old-dependency ratio, the share of old-age relative to the working-age population, isanother commonly adopted measure of population aging in the literature (Ang and Mad-daloni (2005), Krueger and Ludwig (2007), and Takats (2010) among others). Finally,and less frequently, the age of head of household is used in studies with survey data(Bergantino (1998) and Andrews and Sanchez (2011)). Many papers adopt multiple mea-sures of aging, yet in most cases 65 is fixed as a threshold age that distinguishes old-agepeople from the rest of the adult population. This paper contributes to the literature byintroducing a distance to life expectancy as well as the effective retirement age as analternative threshold age that defines old-age population.

Second, this paper finds a significant role of old-age population, specifically the re-tired population, on real house price, given the conventional determinants of house price.Recent empirical studies on the determinants of house prices such as Egert and Mihal-jek (2007), Rae and van den Noord (2006), Hirata et al. (2013) commonly find that realincome, real interest rates, credit growth, demographics, and supply-side factors are im-portant drivers of real house price. However, they do not explicitly consider populationaging as a driver of real house price. On the other hand, using detailed adult age groups,Fortin and Leclerc (2000) shows that the population aged between 25 and 54 played animportant role in real housing price in Canada and predicts that the negative effect of ag-ing, measured by the size of population aged 65 or above, will stay limited in the future,given continuing growth in real income. Similarly, Chen et al. (2012) forecasts real houseprice in Scotland using six adult age-bands (i.e. population aged 25-34, ..., 65-74, 75+),and concludes that demographic change is not an important determinant of house price.Takats (2012) is a recent empirical works on aging and house prices in a global context.Using the old-dependency ratio, it finds that population aging will negatively affect realhouse prices in OECD countries, although the asset price meltdown is unlikely to happen.Consistent with existing studies, this paper predicts a considerable decrease in real houseprice in OECD countries associated with increasing share of population aged 65 or above.However, it also shows that real house price does not necessarily decrease significantly, ifold is defined as effectively retired people.

Lastly, this paper considers a high degree of heterogeneity in population aging amongOECD countries by including Japan and Korea, who are the fastest-aging countries in theworld, in the sample. A number of empirical papers study the effect of aging in a single

7


country, while only a few papers conduct an international analysis. Andrews and Sanchez(2011) examines the relationship between population aging and homeownership in OECDcountries, but they do not include Japan whose population aging is faster than any othercountry. In this paper, I use a country-specific threshold age to define old-age populationso that the analysis can deal with cross-sectional heterogeneity caused by a different paceof population aging.

8


Chapter 3

3 Data and Measurement

3.1 Data Sources

I build a panel dataset of 22 advanced OECD countries - Australia, Austria, Belgium,Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Korea, Nether-lands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom,United States - for the periods between 1970 and 2014. The main variables can begrouped into four categories: house prices, demographics, employment, and other stan-dard macroeconomic variables.

Housing market data is from the OECD House Prices Indicators, in which I use nom-inal house price index and real house price index. Life expectancy at birth, fertility rate,population density and the old-dependency ratio are from the World Bank World De-velopment Indicators (WDI). Population by single year of age for European countries isobtained from Eurostat, which provides population data from age 0 to over 100. For non-European countries, I interpolate population of 5-year age group from the UN PopulationDivision to calculate the single year of age population, following Beer’s methodologyintroduced in NCHS (1999).2

To account for the effect of retirement on housing demand, I use the average effec-tive retirement age from the OECD Ageing and Employment Policies. The conventionalmacroeconomic determinants of house prices such as GDP per capita, 10-year govern-ment bond rate, current account balance, and construction cost are obtained from varioussources including the WDI, the International Financial Statistics, the Jorda-Schularick-Taylor dataset, the OECD, and national statistics departments. Nominal values are alltransformed into real terms using the Consumer Price Index (2010 = 100) from the OECD.

Figure 2 shows the development of demographic factors as well as real house pricesbetween 1970 and 2014 in 22 OECD countries in the sample. A high level of heterogene-ity is present across countries, especially in old-age dependency ratio, the average effec-tive retirement age, and real house price. The obvious and commonly observed trend inthe figure is that the old dependency ratio as well as life expectancy have been increasingover time. The evolution of the effective retirement age and real house price apparentlyvaries wildly across regions: Japan and Korea have a higher effective retirement age with

2In order to minimize any artificial changes in the size of population, I do not combine population dataof a country from different sources. Detailed steps of the interpolation are described in appendix.

9


Figure 2: Demographic changes and real house price in selected OECD countries

60

65

70

75

80

85

Age

1970 1980 1990 2000 2010Year

France Italy

Japan Korea

United Kingdom United States

Life expectancy by country

55

60

65

70

75

Age

1970 1980 1990 2000 2010Year

France Italy

Japan Korea


Effective retirement age by country

10

20

30

40

%

1970 1980 1990 2000 2010Year

France Italy

Japan Korea

United Kingdom Unites States

Old dependency by country

100

200

300

400

500

Index (

1970=

100)

1970 1980 1990 2000 2010Year

France Italy

Japan Korea


Real house price by country

10

Figure 2: Demographic changes and real house price in selected OECD countries

60

65

70

75

80

85

Age

1970 1980 1990 2000 2010Year

France Italy

Japan Korea


Life expectancy by country

55

60

65

70

75

Age

1970 1980 1990 2000 2010Year

France Italy

Japan Korea


Effective retirement age by country

10

20

30

40

%

1970 1980 1990 2000 2010Year

France Italy

Japan Korea

United Kingdom Unites States

Old dependency by country100

200

300

400

500

Index (

1970=

100)

1970 1980 1990 2000 2010Year

France Italy

Japan Korea


Real house price by country

10


Data and Measurement

relatively lower real house price than other countries, while European countries have alower effective retirement age with a boom in real house price in a recent decade.

3.2 Measures of Population Aging

In the literature, population aging is often measured by an increase in the share of old-age population out of either total population or the working-age population, in whichthe threshold age to distinguish old from the rest of adult population is typically fixedas 65. Therefore, the standard measures essentially capture the accumulated stock ofelderly people who are aged over 65. Although these measures are useful to accountfor changes in longevity, they do not allow a possibility that increased life horizon maygenerate different consumption behaviors within the group aged over 65.

Figure 3 shows that there has been an increasing gap between retirement age andlife expectancy in all OECD countries. In most countries, the effective retirement age haslittle increased over time, if not decreased. Therefore, unlike in the 1970s, retirement nowdoes not mean reaching life expectancy in a few years; rather, there is a substantive sizeof retired population who still has relatively long expected remaining life. Given theseobservations, I redefine old in two ways; first, old-age population is considered as peoplewho have a short distance to life expectancy, and second, old-age population is the adultpopulation who is withdrawn from the labor force. By comparing these two measuresalong with the benchmark measure, I try to identify which of two channels of populationaging - short expected remaining life or withdrawal from the labor market - matters moreto real house prices.

I start with the standard representation of three age groups, young, middle, and old,and later I introduce a finer representation of old-age groups, which specify subgroups ofelderly population. The general form of the standard age groups is as follows:

11


Figure 3: Life expectancy at birth and the effective retirement age in selected countries

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

France

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

Italy

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

Japan

50

60

70

80A

ge

1970 1980 1990 2000 2010Year

LE ERA

Korea

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

United Kingdom

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

United States

12

Figure 3: Life expectancy at birth and the effective retirement age in selected countries

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

France

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

Italy

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

Japan

50

60

70

80A

ge

1970 1980 1990 2000 2010Year

LE ERA

Korea

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

United Kingdom

50

60

70

80

Age

1970 1980 1990 2000 2010Year

LE ERA

United States

12



Y 0−19(%) =

( 19∑i=0

agei

Total population

)× 100 (1)

M20−(δ−1)(%) =

((δ−1)∑i=20

agei

Total population

)× 100

Oδ+(%) =

(∞∑i=δ

agei

Total population

)× 100

in which agei is the number of people whose age is i, and δ is a threshold age thatseparates old-age population from the rest of adult population. In the benchmark case δ is65. Young age group (Y 0−19), the share of population aged between 0 and 19, is the onlygroup that is not affected by a threshold age.

In the first alternative definition of old, δ is life expectancy minus ten. This measure ofaging investigates whether a distance to life expectancy plays a significant role in housingdemand by older population. By exclusively capturing the size of people who have max-imum ten years of expected remaining life in any country and year, it tries to account forthe fact that the behavior of people aged over 65 in different time periods may not be nec-essarily the same when their expected life horizon differs. For instance, life expectancyin Korea was 67.9 in 1984, but it surged to 82.15 in 2014, therefore, 65 years old Koreantoday has much longer life horizon than the one would have 30 years ago, which suggeststhat his or her consumption behavior is also likely to have changed during the past threedecades.

δ in the second alternative definition is the effective retirement age, which is the sumof each year of age 40 and over, weighted by the proportion of all withdrawals from thelabor force occurring at that year of age.3 The effective retirement age is different fromthe official retirement age in the sense that it captures when the labor force participationactually ceases on average by the older adult population. Depending on various social andeconomic factors, such as wealth, health, education, social security and pension scheme,

3A detailed explanation on the method for calculating the effective retirement age is available athttp://www.oecd.org/els/emp/39371923.pdf

13


the effective retirement age can be lower or higher than the official retirement age, and italso can change every year. Thus, the second definition of old focuses on to what extentlabor force participation that already accounts for economic and social factors affectshousing demand by older population.

Figure 4: The share of old-age populations in selected countries

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

France

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

Italy

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

Japan

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

Korea

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

United Kingdom

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

United States

14

the effective retirement age can be lower or higher than the official retirement age, and italso can change every year. Thus, the second definition of old focuses on to what extentlabor force participation that already accounts for economic and social factors affectshousing demand by older population.

Figure 4: The share of old-age populations in selected countries

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

France

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

Italy

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

Japan

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

Korea

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

United Kingdom

0

10

20

30

% o

f popula

tion

1970 1980 1990 2000 2010Year

O65+

O(LE−10)+

OERA+

United States

14



Figure 4 presents the share of three different elderly groups in selected countries inthe sample. The solid line represents the share of people aged 65 or above, the benchmarkcase, while the dashed line and the dotted line represent people who have maximum tenyears of expected remaining life and people who are effectively out of the labor market,respectively. Overall, the solid line and the dotted line move closely to each other; theshare of population aged 65 or above has significantly increased over time and so doesthe share of retired population. However, in most countries except Japan and Korea, theshare of retired population is larger than that of the standard old-age population, whichindicates that people withdraw from the labor market before they reach age 65.

To study a finer effect of population aging, I further introduce two subgroups of elderlypopulation - relatively less old (O−) and very old (O+), by setting a second threshold ageδ� which ranges between the first threshold age and life expectancy. δ� is not applied tothe standard old-age population, however, because in some observations life expectancyis lower than age 65. For the alternative cases, δ� is life expectancy minus five, and themean between the effective retirement age and life expectancy, respectively. The generalexpressions of O− and O+ are as follows:

Oδ−(δ′−1)− (%) =

((δ′−1)∑i=δ

agei

Total population

)× 100 (2)

Oδ′++ (%) =

(∞∑i=δ′

agei

Total population

)× 100

in which δ� is defined such that

δ� = life expectancy − 5 (3)

when δ is life expectancy minus ten, and

δ� =δ + life expectancy

2(4)

15


when δ is the effective retirement age.Table 1 provides summary statistics of the main variables based on the observations

in the regressions described in Section 4. From the original dataset that covers 45 yearsof 22 countries, I use five-year interval data as population census is typically conducted infive-year interval. Also, I use a balanced-sample for each regression to compare the effectof the alternatively defined old-age populations, therefore any observations lacking lifeexpectancy and/or the effective retirement age are excluded from the regressions, whichresults in 125 observations. On average, the share of elderly population defined by lifeexpectancy (δ = LE-10) is about 2.5% lower than the share of elderly defined in a standardway (δ = 65), while the share of elderly defined by the effective retirement age (δ = ERA)is 2.5% higher than that. The effectively retired population is the most heterogenousgroup across countries; the maximum share of retired population is 27.7%, whereas theminimum is 5.2% of total population. In both alternative measures of old, the share ofvery old people (O+) is at least 3% larger than that of relatively less old people (O−). Itcan be partially explained by the fact that I include people outlive the life expectancy inO+.

Table 1: Summary statistics

Variable Mean Std. Dev. Min. Max. NReal house price index (log) 4.31 0.35 3.23 5.01 125Real GDP per capita (log) 10.44 0.39 9.00 11.39 172Real construction cost (log) 4.53 0.10 4.15 4.72 125Current account balance (%) 0.19 4.32 -12.36 13.86 125Population density (log) 4.12 1.45 0.90 6.24 125

O (δ = 65) (%) 14.88 2.88 7.69 22.21 125O (δ = LE-10) (%) 12.43 2.43 7.01 17.30 125

O- (δ′ = LE-5) 4.13 0.72 2.67 5.98 125O+ (δ′ = LE-5) 8.30 1.83 3.95 11.77 125

O (δ = ERA) (%) 17.26 4.73 5.22 27.72 125O- (δ′ = (ERA+LE)/2) 7.02 2.57 1.04 13.30 125O+ (δ′ = (ERA+LE)/2) 10.23 2.50 3.89 14.99 125

16


Chapter 4

4 Empirical Analysis

4.1 Empirical Strategy

The main econometric method employed in this study is a panel two-way fixed effectmodel, which removes country- and year-specific unobserved characteristics to accountfor possible omitted variable bias.4 To examine the potential channels through whichpopulation aging affects real house price, I repeat the analysis using three alternativelydefined old-age populations one by one.5 The general expression of the regression modelhas the following form:

RHPit = β0 + β1M20−(δ−1)it + β2O

δ+it + γXit + φi + ηt + εit (5)

in which RHPit represents the the log of real house price index, and Mit and Oit aremiddle and old-age populations, respectively, expressed as a share of total population (%).As discussed in the previous section, three values - 65, life expectancy minus ten, and theeffective retirement age - are assigned to δ. β1 captures the effect of middle-age group(aged from 20 to δ-1) on real house prices and the expected sign is positive since theworking-age population has demand for housing, whereas β2 represents that of the old-age population (aged from δ and over) and the expected sign is negative. By changing thevalue of δ, I compare the sign and magnitude of β1 an β2 in each regression result. Xit isa set of standard explanatory variables for real house price that have been documented inthe literature. It includes the following variables:

• Real GDP per capita: As a consumption good, demand for housing is supposed toincrease as income per person increases. It also captures the effect of business cyclefluctuations on house price. The expected sign of a coefficient is positive.

• Real cost of construction for residential buildings: The cost covers labor, transport,and material costs, and it accounts for the supply side of housing market. Since

4I do not pursue first-difference model as a baseline estimation since the demographic variables are veryslow-moving, which can leave little variation if they were first-differenced. Nevertheless, results from thefirst-difference model are reported in Section 4.3.3 as a robustness check.

5Although it would be nice if the effect of different channels of aging can be simultaneously comparedin the same regression, given that the number of observation is small (125) and the model employs the fixedeffects as well, I only keep essential explanatory variables in the regression to avoid biased results.

17


it is hard to de-trend real house price, which has been consistently increasing insome countries, including the construction cost in the regression aims to captureslow technological growth which has contributed to increasing house price. Theexpected sign of a coefficient is positive.

• Population density: People per square kilometer of land area is included. It capturesboth a change in total population and the land availability. The share of urbanpopulation as well as population growth are also used for robustness checks. Theexpected sign of a coefficient is positive.

• Current account as a share of GDP: Existing studies such as Ferrero (2015) andAdam et al. (2012) show that current account deficit is a fundamental that is closelyrelated to a house price boom. The expected sign of a coefficient is negative.

Additionally, I include country fixed effect (φi) and year fixed effect (ηt) to accountfor country- and year-specific unobserved characteristics.

As a second step, two subgroups of the old-age population - relatively less old popu-lation (O−it) and very old population (O+it) - are included in the regression model:

RHPit = β0 + β1M20−(δ−1)it + β2O−

δ−(δ′−1)it + β3O+

δ′+it + γXit + φi + ηt + εit (6)

in which δ is the same threshold age as those in equation (5), and δ� is the secondthreshold age described in equations (3) and (4). Although the sign of β2 and β3 areboth expected to be negative, possibly different magnitude of the coefficients can suggestwhich stage of old - immediately after entering the old group or at the end of expectedlife - affects more their demand of housing.

Lastly, given that the speed of aging varies among the OECD countries, I test whetherthere is any nonlinear effect of aging on real house prices by including the quadratic termof the old-age population:

RHPit = β0 + β1M20−(δ−1)it + β2O

δ+it + β3O

δ+2

it + γXit + φi + ηt + εit (7)

18


Empirical Analysis

RHPit = β0 + β1M20−(δ−1)it + β2O−

δ−(δ′−1)it + β3O+

δ′+it + β4O

δ−(δ′−1)−

2

it (8)

+β5Oδ′++

2

it+ γXit + φi + ηt + εit

These specifications investigate how the marginal effect of the old-age population onreal house prices changes when the share of old becomes larger than a certain threshold.If the sign of β2 and β3 are different from each other in equation (7), it means the size ofold group has a nonlinear effect on house prices. This nonlinearity test will be also doneon the sub-groups of old-age population as presented in equation (8).

4.2 Results

Before turning to the main regression analysis, I start by running the regression modelin Takats (2010), which examines the relationship between the standard old dependencyratio and real house price in OECD countries, considering real GDP per capita and totalpopulation. For a comparison purpose, I use log-differenced variables in five-year intervaldata with year fixed effects as in Takats (2010).

Results are presented in Table 2 in which δ is the threshold age used in each regres-sion.6 Column (1) uses the old dependency ratio defined in a standard way, and all theexplanatory variables show the expected sign with the coefficients significant at 1 percentlevel, which is comparable and consistent with Takats (2010). When the old dependencyratio is defined by the ratio between effectively retired population and the working agepopulation in column (3), although it still shows a significantly negative coefficient, themagnitude of the coefficient is much smaller than in column (1).7 However, the coeffi-cient for the old dependency ratio in column (2) is not statistically different from zero.Given that the regression is based on the first-differenced variables, the reason could bethat the share of population who has maximum ten years of expected life generally has notchanged much over time in sample countries, as shown in Figure 3, therefore the differ-

6This study covers a longer period than in Takats (2010), but ends up with a smaller number of obser-vation since it focuses on the observations that have information both on life expectancy and the averageeffective retirement age, which is not always available for some countries.

7The working age population here is the population aged between 20 and the effective retirement ageminus one.

19


Table 2: Regression model in Takats (2010)

(1) (2) (3)δ = 65 δ = LE-10 δ = ERA

Real GDP per capita 1.185∗∗∗ 1.177∗∗∗ 1.081∗∗∗

(0.219) (0.228) (0.226)Total population 1.267∗ 1.952∗∗∗ 1.660∗∗

(0.658) (0.699) (0.657)Old-age dependency -0.958∗∗∗ 0.188 -0.332∗∗

(0.273) (0.239) (0.136)Constant -0.0501 -0.119∗∗ -0.0534

(0.0544) (0.0531) (0.0594)Observations 154 154 154Number of countries 22 22 22R2 0.524 0.535 0.522Standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Notes: LE stands for life expectancy, and ERA stands for the effective retirement age.

enced variable may contain little explanatory power, which ends up with the insignificantcoefficient. Nonetheless, the comparison between column (1) and (3) suggests that thedefinition of old age population matters in estimating the negative relationship betweendemographic change and real house prices.

Now, I turn to the main regression models of this paper, equations (5)-(8). Unlikethe regression model in Takats (2010), this regression model uses variables in level infive-year interval data and includes both country and year fixed effects, instead of first-differencing. Since real house price is expressed in logarithm, the coefficient for the agevariables is interpreted as % change of real house price by 1 percent point increase in theage group variables.

Table 3 shows that the non-demographic explanatory variables overall show the ex-pected sign, and among them real GDP per capita and current account are significant inall specifications. In panel A, in which the result of linear regression model reported, thecoefficient for old-age population is negative in all columns. Therefore, a higher shareof elderly population is associated with lower real house price no matter how the elderlypopulation is defined. There can be three possible explanations on this; firstly, as elderlyowners approach to life expectancy, housing supply naturally increases because real es-tates go back to a market, secondly, elderly people no longer purchase a house because it

20


Empirical Analysis

Table 3: Effect of population aging on real house prices

(1) (2) (3) (4) (5)δ = 65 δ = LE-10 δ = ERA

Panel A. Linear effect δ� = LE-5 δ� = (ERA+LE)/2Real GDP per capita 1.272∗∗∗ 1.454∗∗∗ 1.456∗∗∗ 1.359∗∗∗ 1.356∗∗∗

(0.267) (0.308) (0.312) (0.294) (0.242)Real construction cost 0.436∗ 0.416∗∗ 0.424∗∗ 0.599∗∗ 0.377

(0.214) (0.190) (0.194) (0.220) (0.240)Current account -0.0161∗ -0.0182∗∗ -0.0181∗∗ -0.0197∗ -0.0219∗∗

(0.00794) (0.00793) (0.00793) (0.00989) (0.00909)Population density -0.0976 0.271 0.278 0.257 -0.0334

(0.697) (0.693) (0.696) (0.673) (0.694)M 0.0401∗ 0.0177 0.0171 -0.00383 0.00459

(0.0229) (0.0218) (0.0229) (0.0253) (0.0241)O -0.0430∗∗∗ -0.0571∗∗ -0.0246

(0.0138) (0.0210) (0.0252)O- -0.0454 0.00600

(0.0322) (0.0213)O+ -0.0625∗∗ -0.0625∗

(0.0274) (0.0354)Constant -12.17∗∗∗ -13.75∗∗∗ -13.81∗∗∗ -12.90∗∗∗ -11.01∗∗∗

(3.151) (3.434) (3.404) (3.571) (3.527)Observations 125 125 125 125 125Number of countries 22 22 22 22 22R2 0.830 0.816 0.816 0.789 0.801

Panel B. Nonlinear effectO 0.00159 -0.0309 -0.111∗∗∗

(0.0702) (0.129) (0.0382)O2 -0.00155 -0.00103 0.00257∗∗

(0.00242) (0.00435) (0.000966)O- 0.187 -0.0461

(0.228) (0.0452)O+ -0.0440 -0.167∗∗

(0.140) (0.0743)O-2 -0.0277 0.00365

(0.0285) (0.00284)O+2 -0.000950 0.00545

(0.00691) (0.00319)Constant -10.53∗∗∗ -14.04∗∗∗ -14.97∗∗∗ -16.38∗∗∗ -13.88∗∗∗

(3.421) (3.309) (3.228) (3.792) (3.732)Observations 125 125 125 125 125Number of countries 22 22 22 22 22R2 0.831 0.816 0.817 0.813 0.820Robust standard errors, clustered at country-level, are in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

21


is likely that they already own their house, and finally, elderly people may sell their houseand change to a rented one (or a smaller one) after retirement to finance their consump-tion. Another finding in Table 3 is that the magnitude of the coefficient for the old definedby a distance to life expectancy is significantly larger that that of the ones by the standardthreshold age 65 and the effective retirement age. When I use a finer definition of old-agepopulation by introducing second threshold ages in columns (3) and (5), the older elderlypopulation (O+) shows a significantly negative coefficient. Thus, these results overallsuggest that the negative relationship between the share of old and real house price ismore driven by a distance to life expectancy rather than retirement per se. Meanwhile, themiddle age group, who traditionally is considered as home buyers who positively affecthouse prices, does not turn out to significantly affect real house price in any columns.

Panel B of Table 3 tests the existence of nonlinear relationship between old-age pop-ulation and real house prices. Interestingly, nonlinearity is only detected in column (4),in which the old is defined as people completely withdraw from labor force; therefore,as the share of retired people increases, there is a net positive effect on real house prices.The result reflects the fact that retired population is the most heterogenous group in thesample; people retire relatively early in European countries, while that is the opposite inEast Asian countries. Increase in the share of retired population can mean two things;1) either a certain cohort whose absolute size is large starts to retire (i.e. retirement ofthe baby boom generation), or 2) the duration of life after retirement is increased due toextended life span and/or early retirement. One of the possible interpretations of the non-linear effect is that if the share of retired population becomes large enough, either theysimply do not sell their houses as they still have many years to live or they even havedemand for houses for various purposes. The marginal effect of retired population on realhouse price, keeping all the other explanatory variables constant, is visualized in Figure5; real house price continues to decrease until the effectively retired population reachesaround 23% of total population, and then it starts to increase again.

Overall, the results suggest that increase in the share of elderly population is nega-tively associated with real house prices, and the negative relationship is more determinedby a distance to life expectancy rather than retirement. Thus, being retired itself bringsa smaller and smaller negative effect on real house price, given ever increasing life ex-pectancy. Furthermore, it finds the evidence that there is a positive pressure on real houseprice as the share of retired population reaches a certain threshold. Therefore, the resultsimply that effectively retired population matters more to explain real house price than

22


Empirical Analysis

45

67

89

Lin

ear

Pre

dic

tion

0 10 20 30 40 50O

ERA+

Predictive Margins with 95% CIs

Figure 5: Marginal effect of retired population on real house price

population simply aged over 65.

4.3 Robustness Checks

In this section, I conduct a series of checks to confirm the robustness of the differentmagnitude of the negative effect of population aging on housing prices, depending on thedefinition of old.

4.3.1 Robustness to alternative population measures

I first examine whether the results survive when alternative population measures are used.An increase in total population triggers housing demand, thus leads to an increase inreal house prices. To the extent that change in total population is considered, real houseprice will be affected, regardless of aging. Instead of population density used in themain analysis, I alternatively use the share of urban population and total population forrobustness check. In both specifications, old-age population defined by a distance to lifeexpectancy explains the largest part of a decrease in real house prices in the linear model,and the share of retired population shows evidence of the nonlinear effect on real houseprice.

23


Table 4: Robustness check: alternative population variables

(1) (2) (3) (4) (5) (6)Panel A. Linear effect δ = 65 δ = LE-10 δ = ERA δ = 65 δ = LE-10 δ = ERAReal GDP per capita 1.337∗∗∗ 1.572∗∗∗ 1.556∗∗∗ 1.262∗∗∗ 1.436∗∗∗ 1.344∗∗∗

(0.250) (0.284) (0.320) (0.269) (0.314) (0.294)Real construction cost 0.365 0.282 0.392 0.449∗ 0.428∗∗ 0.620∗∗

(0.259) (0.233) (0.265) (0.218) (0.190) (0.226)Current account -0.0146 -0.0185∗ -0.0188∗ -0.0157∗ -0.0179∗∗ -0.0190∗

(0.00967) (0.00927) (0.0105) (0.00819) (0.00811) (0.0101)Urban population -0.0118 -0.0133 -0.0211

(0.0150) (0.0139) (0.0183)Total population -0.0342 0.354 0.381

(0.690) (0.690) (0.645)M 0.0341 0.00999 -0.0119 0.0403∗ 0.0181 -0.00356

(0.0239) (0.0239) (0.0254) (0.0231) (0.0220) (0.0252)O -0.0401∗∗ -0.0557∗∗∗ -0.0258 -0.0422∗∗∗ -0.0560∗∗ -0.0230

(0.0182) (0.0186) (0.0232) (0.0145) (0.0209) (0.0249)Constant -11.72∗∗∗ -11.88∗∗∗ -10.92∗∗ -11.97 -18.54∗ -18.26∗

(3.074) (3.555) (3.979) (10.43) (10.36) (9.668)Observations 125 125 125 125 125 125Number of countries 22 22 22 22 22 22R2 0.833 0.819 0.800 0.830 0.816 0.789

Panel B. Nonlinear effectO -0.0140 -0.0539 -0.0942∗∗ -0.00236 -0.0203 -0.110∗∗∗

(0.0469) (0.117) (0.0406) (0.0717) (0.133) (0.0367)O2 -0.000851 -0.0000727 0.00186∗ -0.00139 -0.00140 0.00258∗∗

(0.00171) (0.00408) (0.000956) (0.00249) (0.00451) (0.000935)Constant -11.26∗∗∗ -11.89∗∗∗ -12.33∗∗∗ -7.789 -19.73∗ -27.48∗∗

(3.280) (3.646) (3.824) (13.26) (10.09) (10.97)Standard explanatory variables Yes Yes Yes Yes Yes YesObservations 125 125 125 125 125 125Number of countries 22 22 22 22 22 22R2 0.834 0.819 0.811 0.830 0.816 0.814Robust standard errors, clustered at country-level, are in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

24


Empirical Analysis

4.3.2 Robustness to old-age dependency ratio

Since the old-dependency ratio is one of the most frequently used measures of populationaging in academic literature as well as in policy discussions, here I use the old-dependencyratio, instead of the share of middle and old age groups. Similar to the main analysis, theworking-age population is defined as the population ages between 15 and δ - 1. Consistentwith the baseline results, the negative relationship seems to be driven by elderly peoplewho have relatively shorter expected remaining life, and the nonlinear effect of old-agepopulation is only significant when the threshold age for old is the effective retirementage.

Table 5: Robustness check: old-age dependency ratio

(1) (2) (3) (4) (5)Panel A. Linear effect δ = 65 δ = LE-10 δ = LE-5 δ = ERA δ = (ERA+LE)/2Real GDP per capita 1.408∗∗∗ 1.517∗∗∗ 1.501∗∗∗ 1.318∗∗∗ 1.343∗∗∗

(0.323) (0.366) (0.378) (0.356) (0.327)Real construction cost 0.365 0.408∗ 0.500∗∗ 0.711∗∗ 0.401

(0.249) (0.230) (0.214) (0.290) (0.271)Current account -0.0207∗∗ -0.0185∗∗ -0.0179∗∗ -0.0170∗ -0.0214∗∗

(0.00797) (0.00812) (0.00859) (0.00960) (0.00930)Population density -0.180 0.319 0.376 0.461 0.0546

(0.756) (0.791) (0.772) (0.761) (0.746)Old-age dependency -0.0316∗∗∗ -0.0332∗∗∗ -0.0518∗∗ -0.00742 -0.0399∗

(0.00768) (0.00982) (0.0188) (0.00611) (0.0200)Constant -10.47∗∗∗ -13.51∗∗∗ -14.12∗∗∗ -14.15∗∗∗ -11.09∗∗∗

(2.878) (3.144) (3.320) (3.503) (3.703)Observations 125 125 125 125 125Number of countries 22 22 22 22 22R2 0.823 0.815 0.812 0.784 0.799

Panel B. Nonlinear effectOld-age dependency -0.0202 -0.0406 -0.0419 -0.0554∗∗ -0.149∗∗∗

(0.0400) (0.0642) (0.110) (0.0213) (0.0524)Old-age dependency2 -0.000237 0.000163 -0.000388 0.000698∗∗ 0.00383∗∗

(0.000822) (0.00125) (0.00373) (0.000284) (0.00162)Constant -9.771∗∗ -13.24∗∗∗ -14.33∗∗∗ -15.63∗∗∗ -12.59∗∗∗

(3.454) (3.173) (3.254) (3.786) (4.128)Standard explanatory variables Yes Yes Yes Yes YesObservations 125 125 125 125 125Number of countries 22 22 22 22 22R2 0.823 0.815 0.812 0.812 0.817Robust standard errors, clustered at country-level, are in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

25


4.3.3 Robustness to first-difference model

To account for possible nonstationarity of the variables such as real house price, realGDP per capita, and population, I repeat the estimation using first-difference model withyear fixed effect in five-year interval data. It is worth emphasizing that population isa very slow moving variable, therefore first-differenced old age variables are likely tolose explanatory power especially when the old is defined as a group with ten years ofexpected remaining life, since there will be only a little change in the share of this groupunless life expectancy and/or fertility rate change dramatically. The results are overallconsistent with the baseline results. Old-age populations are both represented as a shareof total population (col(1)-(3)) as well as old-dependency ratio (col(4)-(6)) in Table 6.Although old-age population based on life expectancy shows no significant effect, theretired population shows evidence of the nonlinear effect on real house price.

26


Empirical Analysis

Table 6: Robustness check: first-difference model

(1) (2) (3) (4) (5) (6)Share of old age population Old dependency ratio

Panel A. Linear effect δ = 65 δ = LE-10 δ = ERA δ = 65 δ = LE-10 δ = ERAΔ Real GDP per capita 1.093∗∗∗ 1.137∗∗∗ 1.091∗∗∗ 1.159∗∗∗ 1.060∗∗∗ 1.031∗∗∗

(0.252) (0.275) (0.306) (0.250) (0.279) (0.295)Δ Real construction cost 0.724∗∗∗ 0.847∗∗∗ 0.799∗∗∗ 0.722∗∗∗ 0.874∗∗∗ 0.792∗∗∗

(0.261) (0.263) (0.260) (0.256) (0.261) (0.266)Δ Current account -0.0156∗∗ -0.0162∗∗ -0.0173∗∗ -0.0148∗∗ -0.0149∗∗ -0.0169∗∗

(0.00644) (0.00686) (0.00679) (0.00625) (0.00689) (0.00693)Δ Population density 1.107∗∗ 1.590∗∗∗ 1.351∗∗∗ 1.290∗∗∗ 1.788∗∗∗ 1.460∗∗∗

(0.470) (0.521) (0.491) (0.498) (0.571) (0.530)Δ M 0.00655 -0.0205 -0.0132

(0.0286) (0.0218) (0.0248)Δ O -0.0424∗∗∗ -0.0121 -0.0239

(0.0150) (0.0287) (0.0194)Δ Old-age dependency -0.593∗∗∗ 0.106 -0.213

(0.186) (0.252) (0.160)Constant -0.0854 -0.110 -0.0933 -0.0944 -0.134∗ -0.0958

(0.0744) (0.0791) (0.0831) (0.0704) (0.0805) (0.0868)Observations 103 103 103 103 103 103Number of countries 22 22 22 22 22 22R2 0.664 0.672 0.669 0.660 0.668 0.661

Panel B. Nonlinear effectΔ O 0.00465 -0.00747 -0.0411∗∗

(0.0311) (0.0327) (0.0192)Δ O2 -0.0251∗∗ 0.00859 0.00879∗∗∗

(0.0121) (0.0161) (0.00308)Δ Old-age dependency 0.0520 0.0537 -0.344∗∗∗

(0.305) (0.286) (0.129)Δ Old-age dependency2 -5.719∗∗∗ -0.270 1.170∗

(1.758) (1.288) (0.678)Constant -0.0990 -0.112 -0.111 -0.104 -0.134∗ -0.114

(0.0774) (0.0799) (0.0801) (0.0704) (0.0788) (0.0848)Standard explanatory variables Yes Yes Yes Yes Yes YesObservations 103 103 103 103 103 103Number of countries 22 22 22 22 22 22R2 0.667 0.675 0.682 0.672 0.666 0.671Robust standard errors, clustered at country-level, are in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

27


Chapter 5

5 Projection under Forward-Looking Scenario

In this section, I investigate the responses of real house price to future demographicchanges based on the regression results. Before I calculate anything, I first check good-ness of fit of the regression models that give significant results in Table 3. The relationshipbetween the log of real house price index (x axis) and the fitted value (y axis) are plottedin Figure 6 for the whole sample period. From the left, the linear benchmark model (δ =65, panel A column (1) in Table 3), the linear life expectancy model (δ = LE-10, panel Acolumn (2) in Table 3), and the nonlinear effective retirement age model (δ = ERA, panelB column (4) Table 3) are presented, respectively. Overall, the models seem to fit the datawell.

33.5

44.5

5R

HP

(F

itte

d)

3 3.5 4 4.5 5RHP (Data)

δ = 65

Col(1) in Panel A

33.5

44.5

5R

HP

(F

itte

d)

3 3.5 4 4.5 5RHP (Data)

δ = LE−10

Col(2) in Panel A

33.5

44.5

5R

HP

(F

itte

d)

3 3.5 4 4.5 5RHP (Data)

δ = ERA

Col(4) in Panel B

Figure 6: Goodness of fit of the regression models in Table 3

5.1 Population Projection

In the next step, I project the share of three alternatively defined old-age population forthe future period following the same way described in equation (1). Firstly, I obtain

28


Projection under Forward-Looking Scenario

population projections for single year of age. All European countries in the sample alongwith the U.S. and Korea are those whose single year of age data is readily available, atfurthest, until 2080.8 Countries whose detailed population projection is not available areexcluded from the prediction.

To calculate the future share of old-age population based on expected remaining life, Iuse the country-specific life expectancy projections suggested by Kontis et al. (2017) andlinearly interpolate and extrapolate the projections until 2055. In the case of the effectiveretirement age, however, there are no projections available. During the sample periodbetween 1970 and 2014, the effective retirement age has changed every year, and it hasnot always moved hand in hand with life expectancy. Thus, I make two assumptions andproject the share of effectively retired population under each scenario: 1) the effectiveretirement age is fixed at the level of 2014, and 2) the effective retirement age increasesover time according to the past change in life expectancy. For the second assumption, Icalculate the average change in life expectancy of each country for the period 1974-2014:the country who has the highest mean change in life expectancy is Korea (average 2.26years of increase), and with the lowest years of average increase is Denmark (average0.84 year of increase), while the overall mean is 1.13 years per five-year.

Figure 7 shows the share of population aged 65 or above (O65+), population with tenyears of expected life (O(LE−10)+), and effectively retired population (OERA+) in selectedcountries. First of all, the share of population aged 65 or above, the solid line, is predictedto increase consistently until around 2050 in all selected countries when it reaches wellabove 30% of total population in fast-aging countries such as Italy and Korea. Meanwhile,in European countries the share of effectively retired population is predicted to be alwayshigher than the share of the standard old-age population, if the retirement age is assumedto be fixed at the current level. This is due to people’s early exit from the labor marketin Europe. In the case of the U.S., the share of standard old-age population and retiredpopulation are projected to be almost the same since the effective retirement age in theU.S. in 2014 is 65.29. In Korea, the situation is the opposite to that of European countries,which means that people exit the labor market much late. If I assume that the effectiveretirement age keeps increasing by 1.13 years in every five-year, however, the share ofretired population is projected to increase only limitedly over time in all countries. Finally,the share of people who have maximum ten years of expected remaining life is projected

8European population projections are taken from the Eurostat, and the other two countries’ data areobtained from national statistics.

29


10

20

30

40

10

20

30

40

2020 2030 2040 2050 2020 2030 2040 2050 2020 2030 2040 2050

France Germany Italy

Korea United Kingdom United States

O65+

O(LE−10)+

OERA+

(ERA fixed) OERA+

(ERA increases)

% o

f popula

tion

Graphs by country

Figure 7: Old-age population projections in selected countries

to either stay at the current level or slightly increase, except in Italy.

5.2 Real House Price Prediction

Since the prediction focuses on the effect of future demographic changes on real houseprice, the values of non-demographic explanatory variables are set to the value of its lastobservation, mostly the observation in 2014, except real GDP per capita and populationdensity, which I can safely assume a continuing change over time. I assume the growthrate of real GDP per capita to be each country’s own average growth rate during thepast ten years, and recalculate population density based on each country’s populationprojection. Then, I predict real house price in selected countries using the three regressionmodels and their corresponding coefficients that I checked the goodness of fit in Figure 6.

In Figure 8, the solid line is real house price predicted by the benchmark regressionmodel (using O65+), while the dashed line and the dotted line are by the life expectancymodel (using O(LE−10)+) and the effective retirement age model (using OERA+), respec-tively. As the size of standard old-age population increases, real house price is predicted to

30



immediately decrease from the level of 2019 in all countries, although the extent to whichit drops varies across countries. However, when I focus on the alternatively defined old-age population, the regression models predict future real house price to be significantlyhigher than the one predicted by the benchmark model in all selected countries. Since theretired population in Figure 8 is calculated under the assumption that the effective retire-ment age is fixed at the level of 2014, the projections suggest that accumulating size ofretired population will put an upward pressure on real house price, therefore house priceswill significantly increase over time. Finally, when old is defined by a distance to lifeexpectancy, the dashed line, real house price is still projected to increase but not as muchas in the case where the old is defined as retired population. Note that the projected levelof real house price varies a lot across countries since the predictions are allowed to beaffected not only by demographic factors but also by each country’s past real GDP percapita growth.

0

100

200

Index (

2019=

100)

2020 2030 2040 2050 2060Year

France

0

100

200

300

Index (

2019=

100)

2020 2030 2040 2050 2060Year

Germany

0

100

200

Index (

2019=

100)

2020 2030 2040 2050 2060Year

Italy

0

100

200

300

400

500

Index (

2019=

100)

2020 2030 2040 2050 2060Year

Korea

0

100

200

Index (

2019=

100)

2020 2030 2040 2050 2060Year

United Kingdom

0

100

200

Index (

2019=

100)

2020 2030 2040 2050 2060Year

United States

RHP65+

RHP(LE−10)+

RHPERA+

(ERA fixed)

Figure 8: Real house price projections in selected countries

Figure 9 shows how the change in the effective retirement age is likely to affect realhouse price. In all selected European countries, the fixed effective retirement age, the

31


0

100

200In

dex (

2019=

100)

2020 2030 2040 2050 2060Year

France

0

100

200

300

Index (

2019=

100)

2020 2030 2040 2050 2060Year

Germany

0

100

200

Index (

2019=

100)

2020 2030 2040 2050 2060Year

Italy

0

100

200

300

400

500

Index (

2019=

100)

2020 2030 2040 2050 2060Year

Korea

0

100

200In

dex (

2019=

100)

2020 2030 2040 2050 2060Year

United Kingdom

0

100

200

Index (

2019=

100)

2020 2030 2040 2050 2060Year

United States

RHP65+

RHPERA+

(ERA increases)

RHPERA+

(ERA fixed)

Figure 9: Real house price projections in selected countries: different assumptions on ERA

32



dotted line, predicts a higher real house price than the one predicted by the increasingeffective retirement age, the dashed line. Since people currently retire early in Europe,raising the effective retirement age means that a large size of retired population who bringsa positive effect on real house price will be reduced or limited over time, which results inlower real house prices. In the U.K and the U.S, however, raising effective retirement agerather predicts a slightly higher increase in real house price than fixed effective retirementage possibly because 1) the share of retired population in these countries is not big enoughin the first place as in other European countries to bring a positive marginal effect onreal house price, and 2) the decreased share of retired population by raising effectiveretirement age reduces the overall negative effect of aging on real house price.

Korea looks exceptional in these figures in the sense that real house price is predictedto increase considerably in the future in any models. As Figure 7 shows, Korea is one ofthe most fast-aging countries in the world, and therefore there is a considerable negativepressure on real house price. However, as mentioned above, the projection assumes thatreal GDP per capita will keep grow at the rate of growth of each country’s own past,which is high enough to cancel out the negative effect of aging in the case of Korea.9

Main findings in this section are as follows. Firstly, under the standard definition ofold, population aging undoubtedly predicts a decrease in real house price. Once old isdefined alternatively, however, increase in the share of old-age group does not necessarilybring a significant drop in real house price. Rather, future real house price can increase,depending on how life expectancy and the effective retirement age change over time.Therefore, a key take-away from this exercise would be that the standard old-age popula-tion does not detect a nonlinear effect of old-age population on real house price and onlypredicts a further decrease in real house price as aging continues, whereas the effectivelyretired population does capture a nonlinear effect and leaves room for policy intervention.In the meantime, although a change in the effective retirement age can be a useful policytool to affect housing market, the effect seems to be highly country-specific even amongthe advanced OECD countries as the speed of aging matters a lot. In European countries,where people currently withdraw from labor market relatively early, raising effective re-tirement age may help avoid an overheated housing market. However, countries such as

9Indeed, when I assume the same real GDP per capita growth in all sample countries, real house price inKorea is estimated to decrease significantly under any regression models. In a similar way, the projectionsof Italy keep decreasing since the growth rate of real GDP per capita is assumed to be negative accordingto its past 10 years of average.

33


the U.K. and the U.S. whose speed of aging is slow and the gap between the effective re-tirement age and the official retirement age is small, longer working lives may not resultin any significant change in real house prices. Last but not least, the negative effect ofaging can be mitigated as long as other macroeconomic variables such as real GDP percapita keep performing well in the future.

34


Chapter 6

6 Conclusion

This paper empirically studies the effect of population aging on real house prices across 22OECD countries over the past 40 years. Specifically, it addresses through which channel,between short expected remaining life and withdrawal from the labor market, populationaging affects real house price more and how the effect can vary if the old-age populationis defined alternatively in a way to reflect different aspects of aging. I first estimate therelationship between the share of population aged over 65 and real house price, givenother standard determinants of real house price. Then, I repeat the regression model usingthe two alternatively defined old-age populations: 1) the population whose distance to lifeexpectancy is maximum ten years and 2) the population who effectively withdrew fromthe labor force.

By introducing the alternative definitions of old-age population, this paper shows thatthe housing demand of people who are traditionally considered old seems to change overtime as expected life horizon increases. Although the old-age population is negativelyassociated with real house price no matter how they are defined, the magnitude of thenegative effect is the largest when the old is defined by a distance to life expectancy,which suggests that the main driver of a negative relationship between aging and realhouse price comes from the later stage of life and not immediately after the age of 65 orretirement. It also shows that the effective retirement age matters more in explaining theeffect of aging on real house price than the age 65, since the share of effectively retiredpopulation has a nonlinear effect on real house price. Relatedly, future real house pricemay not decrease significantly as aging continues. In all sample countries, future realhouse prices based on the regression models are predicted to drop when aging is definedas increasing share of standard old-age population. However, as soon as old is definedas retired population, real house prices are predicted to increase considerably. Finally,the projection exercise suggests that the effect of raising effective retirement age will beheterogenous among OECD countries since the current level of effective retirement aswell as the speed of aging varies wildly across countries.

35


References

References

Abel, A. B. (2003). The Effects of a Baby Boom on Stock Prices and Capital Accumu-lation in the Presence of Social Security. Econometrica 71(2), 551–578.

Adam, K., P. Kuang, and A. Marcet (2012). House Price Booms and the Current Ac-count. NBER Macroeconomics Annual 26(1), 77–122.

Andrews, D. and A. C. Sanchez (2011). The Evolution of ownership Rates in Se-lected OECD Countries: Demographic and Public Policy Influences. OECD Jour-

nal: Economic Studies 2011(1), 1–37.

Ang, A. and A. Maddaloni (2005). Do Demographic Changes Affect Risk Premiums?Evidence from International Data. The Journal of Business 78(1), 341–380.

Bakshi, G. S. and Z. Chen (1994). Baby Boom, Population Aging, and Capital Markets.The Journal of Business 67(2), 165–202.

Battistin, E., A. Brugiavini, E. Rettore, and G. Weber (2009). The Retirement Con-sumption Puzzle: Evidence from a Regression Discontinuity Approach. American

Economic Review 99(5), 2209–26.

Bergantino, S. M. (1998). Life cycle investment behavior, demographics and assetprices. Ph.d. thesis, Massachusetts Institute of Technology.

Bloom, D. E., D. Canning, and G. Fink (2011). Implications of Population Aging forEconomic Growth. NBER Working Papers 16705, National Bureau of EconomicResearch, Inc.

Brooks, R. (2006, July). Demographic Change and Asset Prices. In C. Kent, A. Park,and D. Rees (Eds.), Demography and Financial Markets, RBA Annual ConferenceVolume. Reserve Bank of Australia.

Chen, Y., K. Gibb, C. Leishman, and R. Wright (2012). The Impact of Population Age-ing on House Prices: A Micro-simulation Approach. Scottish Journal of Political

Economy 59(5), 523–542.

Davis, E. P. and C. Li (2003). Demographics And Financial Asset Prices In The Ma-jor Industrial Economies. Public Policy Discussion Papers 03-07, Economics andFinance Section, School of Social Sciences, Brunel University.

36


References

Deaton, A. and C. Paxson (1997). The effects of economic and population growth onnational saving and inequality. Demography 34(1), 97–114.

Ferrero, A. (2015). House Price Booms, Current Account Deficits, and Low InterestRates. Journal of Money, Credit and Banking 47(S1), 261–293.

Fortin, M. and A. Leclerc (2000). Demographic Changes and Real Housing Prices inCanada. Cahiers de recherche 00-06, Departement d’Economique de l’Ecole degestion a l’Universite de Sherbrooke.

Goyal, A. (2004). Demographics, Stock Market Flows, and Stock Returns. Journal of

Financial and Quantitative Analysis 39(01), 115–142.

Greenspan, A. (2003). Aging global population: testimony before the Special Com-mittee on Aging, U.S. Senate, February 27, 2003. Speech 21, Board of Governorsof the Federal Reserve System (U.S.).

Hildebrand, V. (2001). Wealth Accumulation of US Households: What do welearn from the SIPP data? IRISS Working Paper Series 2001-01, IRISS atCEPS/INSTEAD.

Hirata, H., M. A. Kose, C. Otrok, and M. E. Terrones (2013). Global House PriceFluctuations: Synchronization and Determinants. NBER International Seminar on

Macroeconomics 9(1), 119–166.

Kontis, V., J. E. Bennett, C. D. Mathers, G. Li, K. Foreman, and M. Ezzati (2017).Future life expectancy in 35 industrialised countries: projections with a Bayesianmodel ensemble. Lancet 389, 1323–1335.

Krueger, D. and A. Ludwig (2007). On the consequences of demographic change forrates of returns to capital, and the distribution of wealth and welfare. Journal of

Monetary Economics 54(1), 49–87.

Liu, Z. and M. M. Spiegel (2011). Boomer retirement: headwinds for U.S. equitymarkets? FRBSF Economic Letter (aug22).

Mankiw, N. G. and D. N. Weil (1989). The baby boom, the baby bust, and the housingmarket. Regional Science and Urban Economics 19(2), 235–258.

Modigliani, F. and R. Brumberg (1954). Utility analysis and the consumption func-tion: An interpretation of cross-section data. In K. Kurihara (Ed.), Post-Keynesian

Economics. New Brunswick, NJ: Rutgers University Press.

37


Nardi, M. D., E. French, and J. B. Jones (2015). Savings After Retirement: A Survey.NBER Working Papers 21268, National Bureau of Economic Research, Inc.

NCHS (1999). Method for Constructing Complete Annual U.S. Life Tables. Vital andHealth Statistics Series2 No.129, National Center for Health Statistics.

Nishimura, K. and E. Takats (2012). Ageing, property prices and money demand. BISWorking Papers 385, Bank for International Settlements.

Poterba, J. M. (1994). International Comparisons of Household Saving. NBER Books.National Bureau of Economic Research, Inc.

Poterba, J. M. (2001). Demographic Structure And Asset Returns. The Review of Eco-

nomics and Statistics 83(4), 565–584.

Poterba, J. M. (2004). The impact of population aging on financial markets. Proceed-

ings - Economic Policy Symposium - Jackson Hole (Aug), 163–216.

Rae, D. and P. van den Noord (2006). Ireland’s Housing Boom: What has Driven itand Have Prices Overshot? OECD Economics Department Working Papers 492,OECD Publishing.

Takats, E. (2010). Ageing and asset prices. BIS Working Papers 318, Bank for Inter-national Settlements.

Takats, E. (2012). Aging and house prices. Journal of Housing Economics 21(2), 131–141.

Yoo, P. S. (1994). Age dependent portfolio selection. Working Papers 1994-003, Fed-eral Reserve Bank of St. Louis.

Yoon, J.-W., J. Kim, and J. Lee (2014). Impact of Demographic Changes on Infla-tion and the Macroeconomy. IMF Working Papers 14/210, International MonetaryFund.

Egert, B. and D. Mihaljek (2007). Determinants of House Prices in Central and EasternEurope. Comparative Economic Studies 49(3), 367–388.

38


Appendix A

Appendix A Data

I construct a panel data by combining several datasets from different sources. Basically,there are two types of data I use in this study - data for the period 1970-2014 to run themain regression models and data for the period 2014-2055 to project future real houseprice.

First, I use house price dataset from the OECD that includes real house price, nominalhouse price, price to rent ratio, price to income ratio for all 22 OECD countries in thesample. For population data, I use the population of five-year age group by gender fromthe UN Population Division. Total population used in this study is the sum of male andfemale populations. Other demographic variables, which include life expectancy, old-agedependency ratio, population density, urban population, population growth, are obtainedfrom the WDI of the World Bank. The effective retirement age is calculated by the OECDby gender, and I use the average of male and female effective retirement age in the re-gression. I mainly use current account balance data from the WDI, and use the IMF’sInternational Financial Statistics and the Jorda-Schularick-Taylor Macrohistory dataset tocomplement the missing observations. Construction cost is combined from the varioussources. Cost of construction index for residential buildings is obtained from nationalsources for the following countries: France from the Institut national de la statistique etdes etudes economiques, U.K. from the Department for Business, Innovation and Skills,U.S. from the Census Bureau, Korea from the Korean Statistical Information Service(KOSIS), Finland from the Statistics Finland, Australia from the Australian Bureau ofStatistics, New Zealand from the Stats NZ, and Japan from the Statistics Bureau. For therest of the sample countries, I use cost of construction index of residential buildings fromthe OECD and construction cost index of new residential buildings from the Eurostat.

The single-year of age population projections for the U.S. and Korea are obtainedfrom the Census Bureau and the KOSIS, respectively. Otherwise, I use the Eurostat datafor 15 European countries, except Switzerland. Life expectancy projections are takenfrom Kontis et al. (2017), which estimates life expectancy in 2030 for all countries in mysample. I linearly interpolate and extrapolate life expectancy until 2055.

39


Appendix B

Appendix B Interpolation of population data

I interpolate population data of five-year age group using Beer’s formula. Beer’s formulaused in this study has the following general expression:

Px+k = Ck,x−10 5Px−10 + Ck,x−5 5Px−5 + Ck,x 5Px + Ck,x+5 5Px+5 + Ck,x+10 5Px+10

in which Px+k is the population aged x+k (k = 0,1,2,3,4), 5Px is the total population agedx to x + 5, and Ck,x is Beer’s interpolation coefficient. To obtain the size of single-yearaged between 0-4 and 5-9, I use the following formulas:

P0+k = Ck,0 5P0 + Ck,5 5P5 + Ck,10 5P10 + Ck,15 5P15 + Ck,20 5P20

P5+k = Ck,0 5P0 + Ck,5 5P5 + Ck,10 5P10 + Ck,15 5P15 + Ck,20 5P20

Values for Ck,x are taken from Table A in NCHS (1999).Figure 10 compares the share of population aged 65 in New Zealand and the U.S.

from the national statistics and from the interpolation. I use interpolated data to constructthe share of different age populations for countries whose single-year of age populationis not readily available.

.7.8

.91

1.1

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010

New Zealand United States

65 years old (official statistics) 65 years old (interpolated)

% o

f to

tal popula

tion

Year

Graphs by country

Figure 10: Comparison between the official statistics and the interpolated series

40


Appendix C

Appendix C Comparison between male and female

Since in many countries life expectancy as well as the effective retirement age of maleand female population are considerably differ from each other, I further investigate if gen-der differences exist in the effect of population aging on real house prices in Table 6. Irun the same regression models as before but by separating male and female; the old-agepopulation in panel B and C is the share of male elderly out of total male population andthe share of female elderly out of total female population, respectively, calculated basedon the gender-specific life expectancy and effective retirement age. The main differencebetween panel B and C can be found in columns (2) and (3), in which the old age popu-lation is defined by a distance to life expectancy. The effect of female population agingexplains decrease in real house price more than aging of male population does; 1 percentpoint increase in female population who has maximum ten years of expected remaininglife is associated with 6.6 percent decrease in real house price, whereas the same amountincrease in male elderly population is associated with 4.7 percent decrease in real houseprice. The result does not seem to support the belief that male homeownership is generallyhigher than female, thus aging of male population is likely to bring a larger negative effecton real house price than female. Rather, it suggests that since female life expectancy isobviously longer than male, even if there may be a gender gap in homeownership rate,housing demand of the old is still more affected by a distance to life expectancy ratherthan their gender. However, there is no significant difference between the effect of maleand female aging, when aging is measured by an increase in the share of retired popula-tion.

41


Table 7: Comparison between male and female populations

(1) (2) (3) (4) (5)δ = 65 δ = LE-10 δ = ERA

Panel A. Total δ� = LE-5 δ� = (ERA+LE)/2M 0.0329 0.00867 0.00720 -0.0157 -0.00767

(0.0270) (0.0252) (0.0268) (0.0264) (0.0260)O -0.0432∗∗∗ -0.0544∗∗∗ -0.0289

(0.0147) (0.0185) (0.0228)O- -0.0375 -0.00549

(0.0252) (0.0244)O+ -0.0616∗∗ -0.0569∗

(0.0264) (0.0301)R2 0.834 0.825 0.825 0.809 0.816

Panel B. MaleM (m) 0.0139 0.0104 0.00955 -0.0259 -0.0141

(0.0269) (0.0243) (0.0252) (0.0231) (0.0271)O (m) -0.0516∗∗ -0.0478∗∗ -0.0194

(0.0188) (0.0181) (0.0244)O- (m) -0.0288 0.00499

(0.0348) (0.0355)O+ (m) -0.0560∗∗ -0.0445∗∗

(0.0219) (0.0194)R2 0.826 0.825 0.825 0.809 0.816

Panel C. FemaleM (f) 0.0432 0.00896 0.00763 -0.00135 0.00508

(0.0261) (0.0232) (0.0255) (0.0257) (0.0253)O (f) -0.0308∗∗ -0.0668∗∗∗ -0.0340

(0.0136) (0.0198) (0.0208)O- (f) -0.0558∗ -0.0146

(0.0304) (0.0220)O+ (f) -0.0721∗∗ -0.0494∗

(0.0295) (0.0277)R2 0.838 0.836 0.836 0.822 0.825

Standard explanatory variables Yes Yes Yes Yes YesObservations 122 122 122 122 122Number of countries 22 22 22 22 22Robust standard errors, clustered at country-level, are in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

42


Appendix D

Appendix D Additional figures

D.1 Real house price and the share of old: 1970-2014

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Australia

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Austria

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Belgium

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Canada

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Denmark

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Finland

43


0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

France

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Germany

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Greece

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Ireland

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Italy

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Japan

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Korea

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Netherlands

44


Appendix D

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

New Zealand

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Norway

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Portugal

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Spain

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Sweden

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

Switzerland

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

United Kingdom

0

20

40

60

80

% o

f popula

tion

0

20

40

60

80

100

120

140

Index (

2010=

100)

1970 1980 1990 2000 2010

Y0−19

O65+

Real house price

United States

45


D.2 Share of different old-age populations: 1970-20140

10

20

30

01

02

03

00

10

20

30

01

02

03

00

10

20

30

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010

Australia Austria Belgium Canada Denmark

Finland France Germany Greece Ireland

Italy Japan Korea Netherlands New Zealand

Norway Portugal Spain Sweden Switzerland


O65+

O(LE−10)+

OERA+

% o

f popula

tion

Graphs by country

46


Appendix D

D.3 Projections of different old-age populations: 2019-2055

10

20

30

40

10

20

30

40

10

20

30

40

10

20

30

40

2020 2030 2040 2050 2020 2030 2040 2050 2020 2030 2040 2050

2020 2030 2040 2050 2020 2030 2040 2050

Austria Belgium Denmark Finland France

Germany Greece Ireland Italy Korea

Netherlands Norway Portugal Spain Sweden


O65+

O(LE−10)+

OERA+

(ERA fixed)

% o

f popula

tion

Graphs by country

47

www.nbp.pl

Documents

Population aging and housing prices: who are we calling old? · of potential economic effect of population aging, given ever-increasing life expectancy. This paper introduces alternatively